摘要
Intrusiondetectioncanbeessentiallyregardedasaclassificationproblem,namely,dis-tinguishingnormalprofilesfromintrusivebehaviors.Thispaperintroducesboostingclassificationalgorithmintotheareaofintrusiondetectiontolearnattacksignatures.Decisiontreealgorithmisusedassimplebaselearnerofboostingalgorithm.Furthermore,thispaperemploysthePrincipleCom-ponentAnalysis(PCA)approach,aneffectivedatareductionapproach,toextractthekeyattributesetfromtheoriginalhigh-dimensionalnetworktrafficdata.KDDCUP99datasetisusedintheseex-perimentstodemonstratethatboostingalgorithmcangreatlyimprovetheclassificationaccuracyofweaklearnersbycombininganumberofsimple“weaklearners”.Inourexperiments,theerrorrateoftrainingphaseofboostingalgorithmisreducedfrom30.2%to8%after10iterations.Besides,thispaperalsocomparesboostingalgorithmwithSupportVectorMachine(SVM)algorithmandshowsthattheclassificationaccuracyofboostingalgorithmislittlebetterthanSVMalgorithm’s.However,thegeneralizationabilityofSVMalgorithmisbetterthanboostingalgorithm.
出版日期
2007年03月13日（中国期刊网平台首次上网日期，不代表论文的发表时间）
